ON THE GENERALIZED BENJAMIN-ONO EQUATION...Benjamin-Ono to nonlinear Schrödinger equations (or systems) of the form d,u - id2u + P(u,dxu,ü, dxu) = 0 . 1. Introduction The purpose
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
TRANSACTIONS OF THEAMERICAN MATHEMATICAL SOCIETYVolume 342, Number 1, March 1994
ON THE GENERALIZED BENJAMIN-ONO EQUATION
CARLOS E. KENIG, GUSTAVO PONCE, AND LUIS VEGA
Abstract. We study well-posedness of the initial value problem for the gener-
alized Benjamin-Ono equation dtu + ukdxu - dxDxu = 0 , k 6 Z+ , in Sobolev
spaces HS(R). For small data and higher nonlinearities (k > 2) new local
and global (including scattering) results are established. Our method of proof
is quite general. It combines several estimates concerning the associated linear
problem with the contraction principle. Hence it applies to other dispersive
models. In particular, it allows us to extend the results for the generalized
Benjamin-Ono to nonlinear Schrödinger equations (or systems) of the form
d,u - id2u + P(u,dxu,ü, dxu) = 0 .
1. Introduction
The purpose of this paper is to study the initial value problem (IVP) for the
generalized Benjamin-Ono (B-O) equation
dtu + ukdxu-dxDxu = 0, t,x£R, k £ Z+ ,
u(x, 0) = Uo(x),
where Dx = (-d2)1'2.
In the case k = 1 the equation in ( 1.1 ) was deduced by Benjamin [3] and Ono[28] as a model in internal-wave theory. Higher order nonlinearities (specially
the case k = 2) also appear in applications [6]. The generalized B-O equationpresents the interesting fact that the dispersive effect is described by a nonlocal
operator and is weaker than that exhibited by the generalized Korteweg-de Vriesequation
(1.2) dtu + ukdxu + d^u = 0.
Several works have been devoted to the existence problem for solutions of
(1.1) with data u0 £ Hs(R) = ( 1 - d2)~sl2L2(R). In this direction the strongest
results can be gathered in the following theorem.
Theorem 1.1. (i) Let u0 £ HS(R) with s = 0 or s = \. Then for k = 1 ork = 2 there exists a (weak) solution u o/(l.l) such that
u £ L°°(R : Hs) n CW(R : Hs) n L,20C(1 : H^l/2).
Received by the editors July 17, 1991.1991 Mathematics Subject Classification. Primary 35Q30; Secondary 35G25, 35D99.Key words and phrases. Generalized Benjamin-Ono equation, initial value problem, well-posed-
Moreover given V £ (0, T) there exists a neighborhood UUo of Uo in Hs such
that the map uo -» u(t) from UUo to Yt> is continuous. When k = 1 this result
extends to any time interval.
Above we have used the notation:
CW(R : B) = the space of all weakly continuous function
on M to B (Banach space),
andCb(R : B) = C(R : B) n L°°(R : B).
Part (i) of Theorem 1.1 is due to Ginibre and Velo [16]. Part (ii) was proven
by Ginibre and Velo [16] and Tom [40]. Result (iii) was established by Ponce[30]. Finally (iv) follows by combining the works of lorio [18], Abdelouhab, etal. [1], and Ponce [29] (for related results see [32, 21]).
From Theorem 1.1 we see that for k > 2 and u0 £ HS(R) with s < \ theexistence problem for (1.1) is open. On the other hand the proof of the local
well-posedness result for (1.1) in Hs with s > \ (Theorem l.l(iv)) does not
use the dispersive structure of the equation.Also Theorem 1.1 tells us that the smoothing effect of Kato type [20] estab-
lished in solutions of (1.1) is weaker (even locally in time) than that deduced
by Kenig, Ponce, and Vega [23] in solutions of the associated linear problem.
More precisely, if {V(t)}foo denotes the unitary group associated to the linear
problem
d,u-dxDxu = Q, x,t£R,
u(x, 0) = uo(x),
i.e., u(x, t) = V(t)uo(x), then it was shown in [20, Theorem 4.1] that there
exists c > 0 such that for each x £R
aoo \ 1/2\Dx/2V(t)u0(x)\2dt) =c\\u0\\2.
Observe that although the gain of derivatives in ( 1.4) (as in Theorem 1.1) is
also equal to \ , (1.4) describes a L^L2 estimate instead of the L,20CLj20C one
sees in Theorem 1.1.Our main results in this paper show that for k > 2 and small data the results
in Theorem 1.1 can be significantly improved.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
GENERALIZED BENJAMIN-ONO EQUATION 157
Theorem 1.2. Let k>2 and s be such that
s>l ifk = 2,
(1.5) s>¡ ifk = 3,
s>! z/A:>4.
There exists S = ô(k) > 0 such that for any Uo £ HS(R) with ||moIU ,2 < S thereexist T = r(||izo||i,2; k) > 0 (with T(p, k) -> oo as p —► 0) and a unique
(strong) solution u(-) ofthelVP (1.1) satisfying u £ C([-T, T] : HS(R)) and
(1.6) SUP \\D°{S +l/2)u\\Lm-euve < 00o<e<i
wherei/p
IIMIl;l«= / (/r|w(jc'i)|,rfr) dx
and j = min{zc ; 4} .
If Uo £ Hs' (R) with s' > s then the results above hold with s' instead of s
in the same time interval [-T, T].
For T £ (0, T) there exists a neighborhood UUo of uo in HS(R) such that
the map ü0 —» «(Z) from UUo to XST, j is Lipschitz where
XST , = J co : R x [-T, T] -+ | sup \\co(t)\\s 2 < oo[ l-T,T]
and sup ||D^(l+1/2)G>||r,/(1-9),2/e < oo I .O<0<1 x T J
Theorem 1.3. For k > 4 and s > 1 the results of Theorem 1.2 extend to thetime interval (-00,00). In addition, the solution u satisfies
(1.7a) u£Cb(R:Hs(R)),
and
a 00 \ 1/4\\u(.,t)\\ldt) <oo
(1.7c) sup ||Z^(s+1/2)w|U(1-e),2/9<oo.o<e<i * '
(See notations at the end of this section).
As a consequence of Theorem 1.3 we obtain the following scattering type
result (see [36]).
Corollary 1.4. Let k > 4 and uq g HX(R) satisfying the smallness assumption
ofTheorem 1.2. Then for the corresponding solution u(t) of the IVP (\.\) there
exist unique <y0± G Hx (R) such that
(1.8) lim ||m(í)-K(í)ü)o±||i,2 = 0.Í—>±oo
Remarks, (a) Since Theorem 1.2 deals with fractional derivatives in L°° theuse of the one parameter family of norms introduced in (1.6) is necessary, since
D'y is not bounded in L°° and hence complex interpolation is not possible.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
158 C. E. KENIG, GUSTAVO PONCE, AND LUIS VEGA
Notice that for 6= 1 (1.6) provides the sharp version (L^L2.) oftheKato
smoothing effect described in (1.3).
(b) Combining the result (iii) in Theorem 1.1 with those in Theorem 1.3 one
sees that for u0 £ 773/2(R) with 30 = ||"oil3/2,2 < 1 the IVP (1.1) has a globalsolution for power k — 1,4,5,..., j(So). To explain this gap we observe
that in the case k = 1 the global result is due to the conservation laws (see
[5, 9]), and holds for arbitrary data. On the other hand, the result for k > 4
is a consequence of the L4L^°-estimate of Strichartz type [37, 15] satisfied
by solutions of the associated linear problem. In particular (1.7a-c)-(1.8) tell
us that for small data in HX(R) the asymptotic behavior of the solution is
controlled by the dispersive part, and hence scattering occurs.Also worth noting are the works of Bona, Souganidis, and Strauss [7] and
Weinstein [43] where the critical power for stability and instability of the solitary
waves solutions of (1.1) was shown to be k = 2. In the case of the generalizedKorteweg-de Vries both critical values (i.e., that for small data scattering [26]and the one for the stability of the solitary waves [7, 43]) agree and are equal
to 4.(c) The results in Theorem 1.2 could be extended to the limiting cases in (1.4)
(i.e., s = 1 if k = 2 and s = ^ if k = 3) by proving the following estimate:
\l/2
(/:(1.9) / SUP \V(t)Uo\¿dx\ <c||M0||i/2,2.
-00 [-1,1] J
However we do not know whether or not inequality (1.9) is true.(d) Our method of proof combines several estimates concerning the associ-
ated linear problem (1.2) with the contraction principle and the integral equa-
tion form of (1.1)
ÍJo
(1.10) u(t) = V(t)uo- V(t-x)(ukdxu)(T)dx.Jo
The main new tool here is the inhomogeneous version of the smoothing effect
of Kato type (1.3). We shall prove (see Theorem 2.1, estimate (2.5) below) that
\DX /Jo
(1.11) \\DX / V(t-r)F(.,r)dr < C\\F\\LiL2'L2
(see notations at the end of this introduction).
Thus (1.11) affirms that the gain of derivatives in the inhomogeneous case
is twice that described in (1.3) for the homogeneous problem. For further
comments and references related to the estimate (1.11) and its extension to
higher dimensions in the case of the free Schrödinger group see [25].
The same approach presented here (linear estimates and contraction prin-
ciple) applies to other dispersive models. For example, combining the results
below with some obtained in [26], the IVP for the generalized KdV equation
(1.2) is locally well-posed in all HS(R) with
s > I if k = 1,
s > \ if k = 2,
s > ± if k = 3,
s>(k-4)/2k ifk>4.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
GENERALIZED BENJAMIN-ONO EQUATION 159
It easily follows that the IVP
dtu + ukdxu-dxD^u = 0, t, x £ R, « G (1, 2),
u(x, 0) = Uo(x),
is locally well-posed in all HS(R) with s > (9 - 3p)/4 for k = 1 (already
proved in [24, Theorem 1.3]), s > (7-3p.)/4 for k = 2, s> (19-9/*)/12 fork = 3, and so on. In the case studied here p. = 1 the restriction on the size ofthe data appears since the gain of derivatives in ( 1.11 ) is equal to 1 (the amount
of derivative in the nonlinear term). Thus to overcome the loss of derivatives in
the integral equation (1.10) one needs to use the estimate (1.11) complemented
with one for the maximal function (i.e., sup_r<r<r |F(Z)mo|) . However, this
quantity cannot be made arbitrarily small by taking T small. When p > 1 in
(1.12) (higher dispersivity) one just needs to estimate \\V(í)uo\\lp([-t,t]) withp < oo for which the necessary smallness hypothesis holds.
(e) It is easy to see that the group {e"a* J^ satisfies similar estimates to those
discussed in §2 for the group {^(Z)}^ (see [25]). Also our method of proof
works equally well for real or complex valued functions. Therefore the results
in Theorems 1.2-1.3 and Corollary 1.4 extend without major modifications to
the IVP
dtu = id2u + P](u,ü)dxu + P2(u,ü)dxü,
u(x, 0) = uo(x),
where Pj : C2 -» C for j = 1, 2 are polynomials such that
Pj(z\ ,z2)= Yl aJcßz<\z2 •a+ß>2
In particular, when P\(z\, z2) = 2z\, z2 and P2(z\, z2) = z\ we obtain animprovement in the case of small data of the results of Tsutsumi and Fukuda
[41, Theorem 1].It is interesting to remark that when the nonlinearity in (1.13) is replaced by
a polynomial P(u, II, dxu, dxu) with
(1.14) P(z\, z2, z3, z4) = Y, a<*za
|a|>4
the results of Theorem 1.3 and Corollary 1.4 still hold in 7/3(R) and extendto systems. For this general nonlinearity (1.14) local well-posedness for theIVP (1.3) with small data was already established in [25] (in any dimension).
Thus one has the following unusual situation: for small data global existence of
classical solutions can be established, however, for arbitrary data no existence
result is known.
The plan of this paper is as follows: In §2 we deduce the linear estimates
to be used in the proof of our nonlinear result. Section 3 is concerned with
estimates involving fractional derivatives. The proof of these estimates will be
given in [26]. In §4 we shall prove Theorem 1.2. Finally, Theorem 1.3 and
Corollary 1.4 are established in §5.
Notations.
Hf(x) denotes the Hubert transform of /, i.e., Hf(x) = p. v. £ * / =
(zsgn(£)/(£))v.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
160 C. E. KENIG, GUSTAVO PONCE, AND LUIS VEGA
Dx = (-d2)xl2 and D% = HoD$.
Lps(R) = (\-d2)-sl2LP(R).
if^(K) : the space of functions h such that if 0 G C0°°(R) then <j>h £HS(R).
For g:Rx[-T, T] -+ R (or C)
IIsIIl;l* = ( /" (fT \g(x, t)\q dt\ dx
in the case T = oo we shall use || • \\lpl« .
CW(I : B) : the space of all weakly continuous functions on an intervalI into a Banach space B .
Cb(R : B) = C(R : B) n L°°(R : 5).
2. Linear estimates
In this section several estimates concerning the unitary group { ̂ (Z)}^ shall
be deduced. As commented above these will be the main tools in the proof of
the nonlinear results in §4. Thus we consider the linear IVP
d,v - dxDxv = 0, Z, x £ R,
v(x,0) = v0(x),
and
d,w - dxDxw = f(x, t), t, x £R,
w(x,0) = 0,
whose solutions can be described by the group {^(Z)}?0^ , i.e.,
v(x,t) = V(t)v0(x), w(x,t)= [ V(t - x)f(-, t)dxJo
where V(t)Vo(x) = (S¡ * v0)(x) and
/oo
-oo
First we have sharp versions of the smoothing effect of Kato type describedin the introduction.
Theorem 2.1. There exists a constant Co, C\ such that
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
2 \l/2
dx)
1/2
GENERALIZED BENJAMIN-ONO EQUATION 161
Proof. For the proof of (2.3) we refer to [23, Theorem 4.1 ] (see also [24, Lemma
2.1]). For previous estimates of LfQCL20C type we refer to [13, 33, 42].
(2.4) follows from (2.3) by duality.To obtain (2.5) we take Fourier transform in both variables in (2.2). Without
loss of generality we shall assume f £ S(R2). Thus formally we have
Dxw(x, t) = c T r e^e^-^M, x)d^dx.
Using Plancherel's theorem one sees that
(£W. tsfi¡f =,(£|/%«_K¡_/({, „«(/•oo I /-oo 2 \
/ / K(x-y,x)f^(y,x)dy dx)J—oo \J—oo /
with /W denoting the Fourier transform of / in the time variable and K(l, t)the inverse Fourier transform (in the space variable) of the temperate distribu-
tion defined as the principal value of |£|/(t - <j;|<!;|). Hence for x > 0
(2.6) K(l,x) = c ¡™ e^j^r-dn.J-oo i-mm
In a neighborhood of the singular points r\ = 1 and n = ±oo the function
\n\(l - n\r¡\)~x behaves like the kernel of the Hubert transform l/n (or itstranslates) whose Fourier transform is c sgn(y). Thus a comparison argument
shows that for x > 0 K(l, t) is bounded. Since K(l, x) = -K(-l, -x) we
conclude that K £ L°°(RxR) with norm M.Combining this bound with Minkowski's inequality, Plancherel's theorem
and (2.6) it follows that
aoo \ 1/2 / »oo i /-oo 2 \ '\Dxw(x,t)\2dt) =cU y K(x-y,x)f«\y,x)dy dx)
/oo /-oo / /-oo \ 1/2
Wf(t)(y,-)hdy = cM / |/(y,z)|2o-z dy■oo J—oo \J—oo /
which is the desired estimate. However
/OO /»OO 1/ eine^-—/(í,T)</Trfí-00./-00 T — <=KI
may not satisfy the initial condition in (2.2). Thus we need to consider
(2.7) w(x,t) = w(x,t)-V(t)w(x,0)
where
= c r eixi ( ¡°° fW(£, s) sgn(s)eist^ ds] de,J—oo \J — oo /
/oo V(s) sgn(s)/(x, -s) ds.-oo
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
162 C. E. KENIG, GUSTAVO PONCE, AND LUIS VEGA
From (2.4) we infer that
Dlx/2w(x,0)£L2(R),
and by (2.3) that w(x, t) the solution of (2.2) defined in (2.7) satisfies theestimate (2.5). For a more detailed proof we refer to [26].
To interpolate the estimates (2.3)-(2.4) we shall need the following version
of them.
Corollary 2.2.
aoo \ 1/2\Dx/2+iaV(t)v0(x)\2dt) =c0\\v0\\2
for any x £ R with Co independent of a and
(£>,+fa(ísup / \Dx+,a( / V(t-x)f(-,x)dx
2 \ 1/2
dt\
1/2(2.9)
/oo / />oo \ l/¿
(j \f(x,t)\2dt) dx
where c\(a) depends on a but C\(a) - 0(\a\k) (for some k £ N) as \a\ tends
to infinity.
Proof. (2.8) follows directly from (2.3).To prove (2.9) as in (2.6) for x > 0 we consider
A combination of the result 6.11 in [34, p. 51] with the comparison argumentused in the proof of (2.5) shows that Ka(l, t) e L°°(R2) with the norm depend-
ing on a in the appropriate manner. Once this bound has been established the
rest of the proof follows the argument used to obtain (2.5). D
Next we state the generalized one-dimensional version of a Strichartz estimate
in [37] due to Ginibre and Velo [17].
Theorem 2.3. Let p £ [2, oo] and q be such that 2/q = 1/2 - \/p. Then
aoo \ 1/0\\V{t)v0\\$dtj <c||t;o||2
and
( r°° Il r' " \x,q ( f°° • \l/9'(2.11) U |yo V(t-x)f(.,x)dx dt\ <C[J ||/(-,Z)||«,a'zJ
where \/p + \/p' = l/q + \/q' = 1.
Proof. (See [17, p. 377].) G
To complement the previous estimates we need to consider the maximal
function associated with the group {^(Z)}^ i.e., sup, V(t). This idea wasalready used (in a nonoptimal manner) by Ginibre and Tsutsumi [15] in their
work on uniqueness of solutions of the generalized KdV equation.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
I [X sup Dia( fv(t-x)f(-,x)dx) dt)\J-oo -oo<t<oo \Jo / I
< c(a) U°° (y_°° \Dxl2f(x, t)\ dt} dx
1/2
3/4
Proof. For the proof of (2.16) we refer to [23]. D
Interpolating the previous estimates via Stein's theorem on analytic families
of operators [35] we obtain the inequalities to be used in §§4, 5.
Theorem 2.6. For any d £ [0, 1] and T G (0, oo),
(2.17) \\DxW-i)/4V(t)Vo\\L<«<-e)Lye
(2.18)
<c||t>0||2,
*"*({ V(t-x)f(.,x)dx) r4/(l-9).2/9
< cTix-°y2\\Dxl*f\\Lm+e)L, ,
(2.19) D^'2(jj(t-x)f(.,x)dx) <cr 4/(1-9). 2/Í
r 4/(3+9), 2/(2-9)
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
164 C. E. KENIG, GUSTAVO PONCE, AND LUIS VEGA
Moreover the estimates (2.18)—(2.19) still hold with Dx instead of Dx if
one replaces / by Hf in their right-hand sides.
We recall the notation || • IIl'l«
Up
w-(£(/>*. or-*)**)
for Tg(0, oo].
Proof. (2.17) follows from (2.8), (2.15), and the Three Lines Theorem [4].Similarly, (2.18) can be deduced from (2.9), (2.15), and Holder's inequality.
Finally (2.19) can be deduced from (2.9) and (2.16). Notice that for 6 = \,(2.19) is equal to (2.11) with p = q = 6. G
3. Leibniz's formula and the chain rule
In this section we shall state the inequalities needed in §§4 and 5 in the proof
of our nonlinear results. Roughly speaking they are vector valued versions of
the scalar estimates established by Kato and Ponce [22], Christ and Weinstein[10], and Taylor [39] (for previous results in this direction see Strichartz [37,
Theorem 2.1, Chapter 2]). As in those works the proof is based on ideas of
Coifman and Meyer [11] and Bony [8]. In this case we shall combine them
with the Hardy spaces results of Fefferman and Stein [14], the tent spaces of
Coifman, Meyer, and Stein [12] and vector valued inequalities due to Benedeck,
Calderón, and Panzone [2] and Rubio, Ruiz, and Torrea [31].
For the details of the proof we refer to [26].
Weassume /, g : Rx[-7\ T] -> R (or C) with T £ (0, oo] and F : R — R
(or F : C -► C) with F £ CX(R) and F(0) = 0.
Lemma 3.1. Let p, q £ (1, oo) and a G (0, 1). Then for any T G (0, oo]
\\D°(fg)-fDag-gDaf\\LiL,T
{3A) <c\\^f\\^L^^g\\L^
where p,■■, q,■ £ ( 1, oo), i" = 1,2, with
1 J_ 1_ 1 J_ J_P~ Pi P2 ' q ~ q\ Q2 '
and a¡ g [0, a] such that a\ + a2 = a.
The inequality (3.1) still holds in the cases
(3.2a) q = l with a, G (0, a)
and
(3.2b) (p,q) = (1,2).
Moreover in all these cases the estimate (3.1) remains valid with Dx instead of
Dx.
Lemma 3.2. Let p, q £ (1, oo) and a G (0, 1). Then